Chestify

Chestify is an AI-powered engine that transforms chaotic short-form content into a structured learning library. By using Google Gemini to cross-reference claims against trusted sources

Comment

Inspiration We live in an era of infinite scroll but finite attention. Every day, millions of students and developers doomscroll through TikToks and Reels, encountering nuggets of gold—a coding tip, a physics explanation, a history fact—buried in a sea of noise. But two problems persist:

Retention: That valuable video gets lost in the feed instantly.

Misinformation: "Educational" content is often unverified, leading to the spread of false science and fake news.

We asked ourselves: What if we could turn this short-form noise into a permanent, verified treasure chest of knowledge? That’s how Chestify was born. We wanted to build a tool that doesn't just save videos, but actually understands and vets them, contributing to SDG 4: Quality Education.

What it does Chestify is an intelligent video curation engine. When a user pastes a URL (YouTube Short, Reel, or TikTok):

The "Smart Ingest": Our Python backend extracts the transcript and metadata.

The "BS Detector": Google Gemini 1.5 Flash analyzes the claims against trusted sources (Grounding). It flags content as "Verified" (Cyan) or "Misleading" (Orange).

The Treasure Chest: It organizes the content into a beautiful, masonry-style library where users can chat with their videos using an AI RAG pipeline.

How we built it We built Chestify using a hybrid architecture to balance high-performance UI with powerful AI processing:

Frontend: Built with Next.js 14 and Tailwind CSS. We focused heavily on a "Dark Mode Tech" aesthetic, utilizing Framer Motion for the custom "Direction Aware" hover cards and glass morphism overlays.

Backend: A Python FastAPI worker handles the heavy lifting of video extraction (yt_dlp) and serves as the bridge to the AI.

AI Engine: We leveraged Google Gemini 1.5 Flash for its speed and long-context window to process video transcripts and perform fact-checking.

Database: Firebase Firestore allows for real-time status updates (showing the "Processing..." skeleton state turning into a live card instantly).

Challenges we faced The "Glass" Readability Struggle: One of our biggest UI hurdles was the "Direction Aware" hover effect. We wanted a premium frosted glass look, but it kept making the text unreadable over complex video thumbnails. We had to iterate several times to find the perfect balance of backdrop-blur and opacity to ensure accessibility without losing the "Cyber" vibe.

Prompt Engineering for Fact-Checking: Getting the AI to strictly distinguish between "Simplified Explanation" and "Misinformation" was tricky. We had to refine our system prompts to ensure the "BS Detector" was fair but firm.

Real-time State Sync: syncing the Python backend status (processing/error/complete) with the Next.js frontend in real-time required careful state management with Firebase listeners to ensure the user never felt like the app froze.

Accomplishments that we're proud of The "BS Detector": Watching the AI correctly flag a fake health claim about "Alkaline Water" for the first time was a magical moment.

The UI Polish: We didn't just build a tool; we built an experience. From the "Inferno" gradient themes to the fluid animations, it feels like a production-ready SaaS product.

RAG Implementation: Successfully allowing users to "Chat" with their video library showed us the true power of personalized AI education.

Built With

Share this project:

Updates